Scientific state: OBSERVED ≠ CANDIDATE ≠ AI_PREREVIEWED ≠ HI_REVIEWED ≠ VALIDATED ≠ PROPOSED. Read the integrity legend →
iSydney v2.8 · LESSON PASSPORT #001

How can an institution learn from a possible mistake?

This is the first complete learning-loop demonstration. It deliberately stops before inventing a human judgement.

Machine Observation → Candidate Judgement → HI Review → Outcome → Correction / Action → Lesson → Changed Future Rule → Better Observation
1 · MACHINE OBSERVATION

ICIDM-UNASKED-001

pathogen: Cryptococcus · technology: Whole-genome sequencing · observed_publications: 0

Generated question: Why is the Cryptococcus × Whole-genome sequencing relationship sparse in the publication corpus, and is this a true capability gap, a terminology/classification gap, or simply an unrepresented relationship?

Evidence boundary: Publication sparsity does not imply scientific absence or institutional incapability.
2 · MACHINE JUDGEMENT

CANDIDATE_UNASKED — not “true gap”

The machine has only detected a publication-topology signal. It has not established scientific absence, institutional incapability, or research priority.

3 · HI REVIEW PROTOCOL

What must happen before learning?

  1. Inspect representative records that created the sparse/dormant signal.
  2. Check terminology, taxonomy and metadata consistency across time.
  3. Search institutional/public evidence for the relationship outside the current corpus labels.
  4. Ask a domain reviewer whether the relationship should scientifically be expected.
  5. Assign one adjudication outcome and record supporting evidence.
Current real status: PENDING_REVIEW. No outcome has been fabricated for this demonstration.
4 · SIX POSSIBLE REAL OUTCOMES

The same machine signal can mean very different things

MEMORY_RETRIEVAL_GAPCorrectionAdd retrieval/linkage check before future promotion; repair missing road; preserve false-positive lesson.
CLASSIFICATION_GAPCorrectionRepair taxonomy/metadata; rerun detection; record which label produced the illusion.
APPROPRIATE_SPECIALISATIONBoundary learningLearn that sparse is sometimes expected; add domain-expectation gate.
NO_GAPFalse-positive learningRecord why the signal was misleading and reduce recurrence without suppressing rare questions.
CAPABILITY_GAPInstitutional actionConfirm capability history and decide whether restoration, collaboration or acceptance is appropriate.
TRUE_SCIENTIFIC_UNKNOWNResearch opportunityPromote only after evidence review; create a reviewed Question Passport and consider a real experiment.
5 · REALITY RETURN

Cheap checks before expensive experiments

First action: Audit terminology/classification and retrieve representative papers before proposing new experiments.

Stop rule: Do not escalate to a new experiment if the apparent gap disappears after provenance/classification review.

6 · LESSON

The error is not deleted

If review later shows that the candidate was caused by retrieval or classification, iCIDM should preserve what fooled it, repair the relevant road or metadata, and add a pre-promotion check. If the candidate survives as a true scientific unknown, that success also becomes calibration data.

7 · CHANGED FUTURE RULE

Learning means changing future observation

Before: Sparse publication relationship → candidate gap.

After a real correction: Sparse relationship → retrieval check → taxonomy check → domain-expectation check → candidate gap → HI review.

The rule is intentionally conditional until Lesson #001 receives a real adjudication.

PUBLIC AUDIT

Why this matters

A self-learning institution should be able to show not only its current answer, but the history of what it thought, why it changed, what evidence caused the change, and what future behaviour was modified.

← Gap Calibration Lab · iSydney home →